Support Vector Machine model for hERG inhibitory activities based on the integrated hERG database using descriptor...

Area of Science:

• Computational chemistry
• Pharmacology
• Drug discovery

Background:

• Assessing hERG liability is crucial in early drug discovery.
• Machine learning models for hERG inhibition prediction exist but are limited by single-database datasets.
• A need exists for more robust and broadly applicable hERG prediction models.

Purpose of the Study:

• To develop a highly accurate hERG inhibition classification model.
• To leverage a comprehensive dataset integrating multiple public databases.
• To optimize model performance using advanced descriptor selection techniques.

Main Methods:

• Integrated over 291,000 compounds from ChEMBL, GOSTAR, PubChem, and hERGCentral.
• Employed a Support Vector Machine (SVM) classification model.
• Utilized Non-dominated Sorting Genetic Algorithm-II (NSGA-II) for descriptor selection, optimizing for performance and minimal descriptors, alongside ECFP_4 fingerprints.

Main Results:

• Achieved a kappa statistic of 0.733 and an accuracy of 0.984 on the test set.
• The model significantly outperformed existing commercial hERG prediction tools.
• Applicability domain was assessed using molecular similarity analysis.

Conclusions:

• The integrated dataset and optimized SVM model provide superior hERG inhibition prediction.
• This approach enhances the reliability of early-stage drug safety assessments.
• The model offers a valuable tool for predicting hERG liability in drug discovery programs.